(A) Diagrammatic representation of the optimal linear decoder used to reconstruct locomotion speed, showing one example experiment comprising the three putative cell types (Purkinje cells: pPC; Golgi cells: pGC; mossy figres: pMF). Top: above, time course of original speed signal and decoded trace using the entire population; below, decoded speed traces using only homogeneous ensembles. Bottom: normalised firing rate of all units used for the decoder. Right: firing rates of all units, color-coded according to their putative cellular type, are weighted and linearly summed, then half-wave rectified (diode symbol). (B) Decoding performance for all experiments with at least 8 units recorded, (n = 10). (C) For the same 10 experiments used in panel (B), the scores of the best performing ensembles, selected from random samples (see Results), plotted as a function of the number of units for the selected experiments. For each ensemble size, four different measures are shown: the performance of heterogeneous populations (grey) and of homogeneous ensembles (colour-coded as in panel A).